Base URL

https://api.cognitivers.com/v1        chat, embeddings, model list
https://stream.cognitivers.com/v1     the same API, for very large inputs
https://api.cognitivers.com/health    public health probe

Paths are appended to the base URL, so a chat request goes to https://api.cognitivers.com/v1/chat/completions.

Which host to use

Use api.cognitivers.com by default. Use stream.cognitivers.com when the input is very large, in the hundreds of thousands of tokens, because reading a long prompt takes time before the model writes its first token. The streaming host holds that connection differently so a long read is not treated as a dead one. Same key, same request body, same response shape.

You can switch hosts per request without changing anything else. A pipeline that normally talks to api. can send one oversized job to stream. and keep going.

What is public

Only /v1/* and /health. Everything else on those hosts, including operational endpoints, is not reachable from the internet. The health probe is public on purpose, so you can build a status check without credentials:

curl https://api.cognitivers.com/health
# {"status":"ok"}

Discovery

GET /v1/models lists the model ids your key can call. Clients that support discovery, such as GitHub Copilot in VS Code, can read this instead of hardcoding a list. See List models.